Games interlocutors play: New adventures in compositionality and conversational implicature
How much of linguistic meaning is simply a corollary of rational behavior? And what is the best way to represent and compute conversational implicatures? I aim to show that the Games of Partial Information (GPIs) of Parikh (2001) are descriptively and explanatorily superior to the leading pragmatic theories of conversational implicature. Explicitly based on utility maximization, Parikh's game-theoretic account of communication captures the distinct advantages of the current disparate theories of conversational implicature. It has the clarity, focus, and explicit predictions of the neo-Griceans' accounts of utterance meaning (Levinson, 2000), while also demonstrating the flexibility of relevance theorists' accounts focused on utterance interpretation (Sperber and Wilson, 1995). Bidirectional Optimality Theory (Dekker and van Rooy, 2000), another game-theoretic/utility-based framework, is a marked improvement over neo-Gricean and relevance theory, but it lacks the scope and power needed to fully account for the context sensitivity of conversational implicature. The prior probabilities of GPIs (and their weighted sums) allow contextual factors to be taken into account, and can explain hitherto unexamined aspects of scalar implicature. It is shown that in some cases both Sauerland's and Chierchia's theories fail and in other cases one fails and the other does not. GPIs provide a framework that obviates the need for exclusively localist or globalist accounts of implicature. The insight missing from Chierchia's and Sauerland's theories is that their sets of predictions are not mutually exclusive: whether utterances have Chierchia-style localist meanings or a Sauerland-style globalist meanings is not a question that can be settled across the board, once and for all. Rather, whether a particular utterance has a localist or globalist meaning depends on a set of context-sensitive parameters: prior probabilities of possible intended meanings, the utterance costs of unambiguous expressions of those meanings, and the (dis)utility that the interlocutors derive from (mis)communicating those meanings. In cases for which the localist and the globalist make the same predictions, there is simply no fact of the matter of whether that meaning was local or global---the problem is ultimately solved on a higher level of abstraction, rendering localism and globalism both epiphenomenal.
0984: Computer science